动脉瘤性蛛网膜下腔出血患者围手术期内并发医院感染风险列线图模型的构建与验证

Construction and validation of a nomogram model of nosocomial infection risk in patients with aneurysmal subarachnoid hemorrhage during perioperative period

目的调查动脉瘤性蛛网膜下腔出血(aSAH)患者围手术期内并发医院感染的相关因素,构建能够准确预测医院感染发生风险的列线图模型。方法采用回顾性分析的方法收集2019年1月至2021年6月在湖南省常德市第一人民医院神经外科接受手术治疗的aSAH患者的临床资料,按照3∶1的比例随机分为试验队列与检验队列。将试验队列按患者是否感染分为感染组与未感染组,并用于进行单因素分析、多因素Logistics回归分析以及列线图的构建(列线图的构建采用RStudio软件的rms程序包进行)。构建的列线图分别在试验队列与检验队列进行验证,其中采用受试者工作特征曲线(ROC)分析验证模型区分度[曲线下面积(AUC)为0.5~0.7时有较低准确性,AUC为>0.7~0.9时有一定准确性,AUC为0.9以上时有较高准确性];校准曲线(采用Brier值评估模型准确性,范围为0~1分,0~<0.25分表示模型具有预测价值,Brier得分越接近0表示模型准确度越高;采用平均绝对误差评估模型实际预测与校正后预测的一致性,平均绝对误差越接近0表示一致性越好)与Hosmer-Lemeshow检验(P>0.05表明拟合方程与真实方程基本无偏差,P值越大越好)判断模型校准度;决策曲线分析验证模型的临床有效性(决策曲线中模型的获益率高于所有患者发生医院感染并接受治疗或所有患者不发生医院感染并均不治疗,表明模型有实用性或有效)。结果共纳入aSAH患者372例,其中试验队列279例,检验队列93例;发生医院感染114例(30.65%),其中呼吸道感染59例(51.75%),泌尿系统感染18例(15.79%),中枢神经系统感染13例(11.40%,其中颅内感染9例,脑膜炎2例,无脑膜炎性椎管内脓肿2例),其他感染24例(分别为血液系统感染10例,气管炎或支气管炎8例,心血管系统感染6例)。279例试验队列患者中,未发生感染195例,发生感染84例,其中呼吸道感染49例,泌尿系感染16例,中枢神经系统感染11例,其他感染8例(分别为血液系统感染4例,呼吸系统除呼吸道外的其他感染2例,心血管系统感染2例)。多因素分析结果显示,合并高血压病(OR=2.175,95%CI:1.058~4.475)、合并糖尿病(OR=5.766,95%CI:2.247~14.769)、责任动脉瘤位于后循环(OR=2.404,95%CI:1.146~5.044)、手术时间长(OR=1.053,95%CI:1.030~1.077)、格拉斯哥昏迷量表评分低(OR=0.507,95%CI:0.404~0.638)是aSAH患者围手术期并发医院感染的独立危险因素(均P<0.05)。风险列线图模型在试验队列与检验队列的ROC AUC分别为0.847(95%CI:0.796~0.899)与0.862(95%CI:0.799~0.932),模型区分度良好;校准曲线中试验队队列与检验队列的Brier值分别为0.136与0.133分,拟合优度检验得出χ^(2)=8.242,P=0.410,提示模型校准度较高,平均绝对误差分别为0.017与0.037,提示预测概率与实际概率具有良好的一致性。决策曲线中阈概率值设定为32.0%时,两队列人群临床净获益率分别为60.0%和65.0%,表明模型具有临床有效性。结论基于Logistic回归分析建立的aSAH患者围手术期并发医院感染预测模型具有理论与实际双重价值,可为针对性做好医院感染防控提供指导。

Objective To investigate the related factors of nosocomial infection in patients with aneurysmal subarachnoid hemorrhage(aSAH)during the perioperative period and construct a nomogram model that can accurately predict the risk of nosocomial infection.Methods The clinical data of aSAH patients who underwent surgery in the Department of Neurosurgery,the First People′s Hospital of Changde City from January 2019 to June 2021 were retrospectively collected and analyzed.Patients were randomly divided into experimental cohort and test cohort according to a ratio of 3∶1.The experimental cohort was divided into infected group and non-infected group according to whether the patients were infected,and was used for further univariate analysis,multivariate Logistic regression analysis and the construction of nomogram(the nomogram was constructed using the RMS package of RStudio software).The constructed nomogram was verified in the experimental cohort and the test cohort respectively,and receiver operating characteristic curve(ROC)analysis was used to verify the discrimination of the model(low accuracy with area under the curve[AUC]0.5-0.7;certain accuracy with AUC>0.7-0.9;high accuracy with AUC above 0.9).Calibration curve(the Brier value is used to evaluate the accuracy of the model,ranging from 0 to 1 point.0-<0.25 points indicate that the model has predictive value,and the closer the Brier score is to 0,the higher the accuracy of the model;The average absolute error is used to evaluate the consistency between the actual prediction of the model and the corrected prediction;The closer the average absolute error is to 0,the better the consistency)and Hosmer-Lemeshow test(P>0.05 indicates that there is basically no deviation between the fitted equation and the real equation;the larger the P value,the better)were used to judge the degree of model calibration.The decision curve verifies the clinical validity of the model(When the benefit rate of the model in the decision curve is higher than that of all patients with nosocomial infection and receiving treatment or all patients without nosocomial infection and receiving no treatment,it indicates that the model is practical or effective).Results A total of 372 patients with aSAH were enrolled,including 279 in the experimental cohort and 93 in the test cohort.Nosocomial infections occurred in 114 cases(30.65%),of which 59 cases were respiratory infections(51.75%),18 cases were urinary system infections(15.79%),13 cases were central nervous system infections(11.40%,including 9 cases of intracranial infection,2 cases of meningitis,2 cases of intraspinal abscess without meningitis),and 24 cases were other infections(10 cases of blood system infections,8 cases of tracheitis or bronchitis,and 6 cases of cardiovascular system infections).Among the 279 patients in the trial cohort,195 had no infections and 84 had infections,including 49 respiratory tract infections,16 urinary tract infections,11 central nervous system infections,and 8 other infections(4 cases of blood system infections,2 cases of respiratory system infections other than respiratory tract,2 cases of cardiovascular system infections).Multivariate analysis showed that hypertension(OR,2.175,95%CI 1.058-4.475),diabetes(OR,5.766,95%CI 2.247-14.769),responsible aneurysm located in the posterior circulation(OR,2.404,95%CI 1.146-5.044),long operation time(OR,1.053,95%CI 1.030-1.077),and low Glasgow Coma Scale score(OR,0.507,95%CI 0.404-0.638)were independent risk factors for perioperative nosocomial infection in patients with aSAH(all P<0.05).The AUC of ROC of the risk nomogram model in the experimental cohort and the test cohort were 0.847(95%CI 0.796-0.899)and 0.862(95%CI 0.799-0.932),respectively,and the model distinguishability was good.The Brier values of the experimental team cohort and the test cohort in the calibration curve was 0.136 and 0.133,respectively,and the goodness of fit test shows χ^(2)=8.242,P=0.410,which indicates that the model has a high degree of calibration.The mean absolute errors were 0.017 and 0.037,respectively,indicating that the predicted probability is in good agreement with the actual probability.The threshold probability value in the decision curve is set to 32.0%,and the net clinical benefit rates of the two cohorts are 60.0%and 65.0%,respectively,indicating that the model has clinical validity.Conclusion The predictive model of perioperative nosocomial infection in patients with aSAH based on Logistic regression analysis has both theoretical and practical values and can provide guidance for the prevention and control of nosocomial infections.